X-ray properties of UV-selected star forming galaxies at z~1 in the Hubble Deep Field North

0.43 MB·English

by  E. S. Laird

Mon.Not.R.Astron.Soc.000,1–11(2004) Printed2February2008 (MNLATEXstylefilev2.2)
X-ray properties of UV-selected star forming galaxies at
z ∼ 1 in the Hubble Deep Field North
E. S. Laird,1⋆ K. Nandra,1 K. L. Adelberger,2 C. C. Steidel,3 N. A. Reddy3
1Astrophysics Group, Imperial College London, Blackett Laboratory, Prince Consort Road, London SW7 2AZ, UK
5 2Carnegie Observatories, 813 Santa Barbara Street, Pasadena, CA 91101, USA
0 3California Institute of Technolgy, MS 105-24, Pasadena, CA 91125, USA
0
2

n
a Accepted January2005.
J

9
1 ABSTRACT
We present an analysis of the X-ray emission from a large sample of ultraviolet (UV)

1 selected, star forming galaxies with 0.74 < z < 1.32 in the Hubble Deep Field North
v
(HDF-N) region. By excluding all sources with significant detected X-ray emission in
1
the 2 Ms Chandra observation we are able to examine the properties of galaxies for
1
which the dominant emission in both UV and X-ray is expected to be predominantly
4
due to star formation. Stacking the X-ray flux from 216 galaxies in the soft and hard
1
bandsproducessignificantdetections(14.9σand3.2σ,respectively).Thederivedmean
0
5 2–10keVrest-frameluminosityis2.97±0.26×1040ergs−1,correspondingtoanX-ray
0 derived star formation rate (SFR) of 6.0±0.6 M⊙ yr−1. Comparing the X-ray value
/ withthemeanUVderivedSFR,uncorrectedforattenuation,wefindthattheaverage
h
UV attenuation correction factor is ∼3. By binning the galaxy sample according to
p
UV magnitude and colour, and stacking the observed frame soft band X-ray flux in
-
o eachbin,correlationsbetweenUVandX-rayemissionareexamined.Wefindastrong
r positive correlation between X-ray emission and rest-frame UV emission, consistent
t
s with a strict linear relationship, LX ∝ LUV, at the 90 per cent level. A correlation
a
between the ratio of X-ray-to-UV emission and UV colour is also seen, such that
:
v LX/LUV increases for redder galaxies. We find no direct relation between X-ray flux
i andUV colour.GiventhatX-rayemissionoffersaview ofstarformationregionsthat
X
is relatively unaffected by extinction, results such as these can be used to evaluate
r the effects of dust on the UV emission from high-z galaxies. For instance, using the
a
observedcorrelationbetweenUVcolourexcessandtheratioofX-ray-to-UVemission–
ameasureofUVobscuration–wederivearelationshipforestimatingUVattenuation
correctionsasafunctionofcolourexcess.Theobservedrelationisinconsistentwiththe
Calzetti et al. (2000) reddening law which over predicts the range in UV attenuation
corrections by a factor of ∼100 for the UV selected z∼1 galaxies in this sample.
Key words: galaxies: starburst – galaxies: high-redshift – X-rays: galaxies
1 INTRODUCTION identified in these surveys are observed during the peak
of the cosmic star formation rate density (e.g Lilly et al.
The vast majority of known high redshift galaxies are se-
1996; Madau et al. 1996; Hugheset al. 1998) and represent
lected in the rest-frame ultraviolet (UV). For instance the
asignificantfraction ofthetotalstarformation athighred-
Lyman break technique, which selects actively star form-
shift (Adelberger & Steidel 2000). The star formation rates
ing galaxies based on their broad band ultraviolet (UV)
(SFRs) derived from these UV-selected galaxies are there-
colours, is the primary method for efficiently selecting red-
forecrucialfordeterminingthehistoryofstarformationand
shift z&3 galaxies (e.g. Steidel et al. 1996; Steidel et al.
heavyelement production in the Universe.
1999; Steidelet al. 2003) and has inspired similar tech-
niques to isolate star forming galaxies from 1 . z . 3 However the SFRs obtained from the rest-frame
(Adelberger et al. 2004; Steidel et al. 2004). The galaxies UV fluxes are very sensitive to the effects of dust
attenuation intrinsic to the galaxies (Kennicutt 1998).
Correcting the observed fluxes for the attenuation re-
⋆ E-mail:[email protected]; quires an understanding about the true nature of the
2 E. S. Laird et al.
sources and their dust content that is currently lack- (2004). Briefly, the BBGs have been selected based on the
ing. To overcome this, average corrections are applied strengthsoftheBalmerand4000 ˚Abreaksintheirspectral
based on dust attenuation laws observed in local starburst energydistributions,designedtoidentifystarburst,Sb-and
galaxies (e.g. Calzetti, Kinney,& Storchi-Bergmann 1994; Sc-typestar forming galaxies (Adelberger et al. 2004). The
Calzetti et al. 2000; Meurer, Heckman,& Calzetti 1999, candidate list, which includes all BBGs with spectroscopi-
hereafter MHC99), which appear to be the local analogues cally confirmed redshifts with 0.74 < z < 1.32, is consid-
ofthehighredshiftLBGs(e.g.Steidel et al.1996;Giavalisco erably larger than that used in N02 thanks to the exten-
2002).Howeverthereisconsiderabledisagreementaboutthe sive spectroscopic effort of the Team Keck Treasury Red-
sizeofthecorrectionstobeappliedwithsomeevidencethat shift Survey in the Great Observatories Origins Deep Sur-
attenuation has been overestimated (Bell 2002). vey(GOODS)-Northfield(Wirth et al.2004).Thereare255
X-rayobservationscanaddtothisdebatebyproviding galaxiesinourCDF-NBBGsample,withz=0.95.Un,G,R
a view of galaxies that is relatively unaffected by dust and and I magnitudes and colours have been measured for all
extinction.HardX-raysareemittedfromstarforminggalax- theobjects.
iesbyacombinationofhighmassX-raybinaries(HMXBs),
supernovae remnants and hot gas (Fabbiano 1989). Lo-
callystrongcorrelationsbetweenX-rayluminosityandother 2.2 X-ray data
SFR measures have been seen (David,Jones, & Forman
TheHDF-NregionhasbeenobservedwithChandra ACIS-I
1992), showing that X-ray luminosity can be used
20timessincelaunch,withthetotalexposuretimeapprox-
as a SFR measure (Ranalli, Comastri, & Setti 2003;
imately 2 Ms. Analysis of these observations, and a point
Grimm, Gilfanov, & Sunyaev2003;Nandraet al.2002).By
source catalogue, was presented in Alexander et al. (2003;
comparing X-ray derived SFRs with UV estimates we can
hereafterA03).Forthispaperweusedtherawdataavailable
independently check the validity of the attenuation correc-
from thepublic archive.
tions and gain insight into the nature and dust content of
The HDF-N was observed over a period of 27 months
theUV-selected galaxies in the early Universe.
during which time the observing conditions and operating
Although the majority of high redshift star forming
mode of the telescope changed. The 20 observations can
galaxies are not bright enough to bedetected with thecur-
broadlybeseparatedintothreegroups:thefirstthreeobser-
rent class of X-ray telescopes, with deep Chandra obser-
vationsweretakeninFaintmodewithafocalplanetemper-
vations the average properties of a sample can be deter-
◦
ature of 110 C, the following 9 observations also in Faint
mined using stacking techniques (e.g. Hornschemeier et al. − ◦
mode with temperature 120 C, and the final 8 observa-
2001; Brandt et al. 2001; Nandraet al. 2002). Several ef- −
tions, constituting the second 1 Ms, taken in Very Faint
forts at this have already been made. Brandt et al. (2001)
◦
mode at -120 C. The differences in the observing modes
stacked theemission from 24 z 3 LBGs in the 1 Ms Chan-
∼ at times warranted separate treatment when reducing the
draDeepField-North(CDF-N),detectingasignificant 0.5–
data.
2keVobserved-framesignal.Usinglocalstarburstreddening
′
Our analysis is restricted to an approximately 10.3
relations Seibert, Heckman,& Meurer (2002) predicted the ′ ×
10.3 region within the larger ACIS-I field of view. This en-
mean X-ray luminosity of the24 LBGs and found it agreed
′ ′
compasses theoptical BBG surveyregion (8.7 8.7).
well with the stacking results. Also in the 1 Ms CDF-N, ×
Nandraet al. (2002) analysed the emission from two much
larger samples of 148 z 3 LBGs and 95 z 1 star forming
∼ ∼ 2.3 Data reduction
galaxies.Morerecentlyusingthe2MsdataintheGOODS-
North field, Reddy& Steidel (2004) analysed the stacked Data reduction was carried out using the Chandra X-ray
emission from a sample of UV-selected z 2 galaxies and ? Center (CXC) Chandra Interactive Analysis of Observa-
∼
analysedtheemission fromlargesamplesofz 3,4,5and6 tions (CIAO) software, version 3.0.1, and the Chandra cal-
≃
LBGs. ibration database (CALDB) version 2.23. Using the tool
In this paper we seek to add to this work by improv- ACIS PROCESSEVENTS, all observations had up-to-date
ing uponand extendingtheanalysis of the z 1galaxies by gain maps and geometry solutions applied. In addition, the
∼
Nandra et al. (2002; hereafter N02) using the 2 Ms obser- observations taken at -120◦ C were corrected for radiation
vation of theHDF-Nregion. Using amuchlarger catalogue damage inflicted in the first few months of Chandra op-
of 255 z 1 UV-selected Balmer break galaxies (BBGs) we erations using the standard CXC CTI correction, and the
∼
assess the X-ray properties and the relations between UV eight observations taken in Very Faint mode were further
and X-ray emission of the galaxies, during a period where cleaned by identifying likely background events. The event
the star formation density of the Universe was at its peak. files were filtered using standard screening criteria – ASCA
Throughout, we assume a standard, flat ΛCDM cosmology eventgrades0,2,3,4,6, Good TimeIntervalsandstatus=0–
with ΩΛ =0.7 and H0 =70 km s−1 Mpc−1. to remove bad and flaring pixels, bad columns, cosmic ray
afterglows etc. Source subtracted lightcurves were created
and periods where the background was more than 3σ from
themeanvaluewereflaggedusingANALYZELTCRV.Allthe
2 DATA AND ANALYSIS periods of high background identified by ANALYZELTCRV
wereremovedinaddition toafurther44.4ksfrom observa-
2.1 Optical data
tion 2344 where theprocedure failed to identify a flare and
The BBG candidates were colour selected from optical im- 15.2ksfrom observation3389whereitfailed toidentifythe
agesusingphotometriccriteria,detailedinAdelberger et al. full duration of a flare.
X-ray properties of star forming galaxies at z ∼1 3
theMKPSFtoolwiththeCALDBPSFlibraryfileacisi1998-
11-052dpsf1N0002.fits to create an image of the PSF at the
corresponding detector position in each of the observations
andthencombinedtheimages,weightedbyexposure,tocre-
ateanaveragePSFimage.AradialprofileofthePSFimage
was then obtained by extracting counts from 30 concentric
circlesouttoamaximumradiusof30arcseconds.Assuming
all counts are contained within the 30 arcsecond radius we
thencalculatedbyinterpolationtheradiusatwhich,say,90
percentofthecountsareenclosed(referredtoasthe90per
centencircledenergyfraction,EEF)andalsothefractionof
counts enclosed within a given fixed radius, say, 1.5 arcsec-
onds2. This method can be repeated for different energies,
positions, EEFs and radii.
Figure 1. The BBG survey area versus the minimum effective Finally,PSF-corrected countratesshould beconverted
exposure in the soft-band exposure map. The dashed lineshows to flux. For this conversion we have assumed a power-law
the solidangle of the regionof our analysis. Over 98per cent of source spectrum with Γ=2 in the all bands and Galactic
the BBG area has at least 1 Ms effective exposure and over 85 column density N = 1.6 1020 cm−2 (Stark et al. 1992).
H
percenthasatleast1.5Ms. ×
Whenconvertingfromcountstofluxthefull,hardandultra-
hard bands were extrapolated to the standard upper-limit
of 10 keV.
Before coadding theeventfiles, therelative astrometry IthasbeenfoundthattheChandra ACISquantumeffi-
wasimprovedbyregisteringeachobservation tothecoordi- ciency(QE)hasbeensufferingcontinuousdegradationsince
nate frame of observation 1671 using bright sources within launch – most likely a result of the build-up of material on
6 arcminutes of the aim points. This was done using the theACISdetectorsoropticalblockingfilers(Marshall et al.
ALIGN EVT tool written by T. Aldcroft1. The root-mean- 2004) – which affects the spectral response of the detec-
squareshiftappliedtotheobservationswas0.54arcseconds tors and, therefore, the counts to flux conversion. We have
in RA and 0.38 arcseconds in Dec. The average aim point calculated the necessary corrections to the fluxes using the
for the combined observations, weighted by exposure time, ACISABSmodel,version 1.1-23,4.ForGalactic columnden-
was α = 12h36m47s.59, δ = +62◦14′08′.′1 (J2000.0) and the sityandΓ=2thesoft,hard,fullandultra-hardfluxesshould
total exposure time was 1.86 Ms. beincreasedby14.7,0.6,11.5and0.1percentrespectively.
We constructed event files and images in four bands: All fluxes quoted in this paper have been corrected for the
0.5–2keV(softband),0.5–7keV(fullband),2–7keV(hard QEdegradation.
band) and 4–7 keV (ultra-hard band). Effective exposure
mapswereconstructedusingtheMERGEALLtoolforeach
of the four bands – these take into account the effects of 2.4 Source detection
vignetting, gaps between chips, bad column and bad pixel
Sourcedetectionwascarriedoutinthefield,thereasonsfor
filtering and are required for converting observed counts to
which are threefold: (1) identify BBG galaxies that are di-
flux. The maps were created at a single energy representa-
rectly detected at 2 Ms, (2) use these objects to determine
tive of the photons in each band: 1 keV for the soft, 2.5
any overall offsets between the coordinate reference frames
keV for the full, 4 keV for the hard and 5.5 keV for the of theoptical and X-ray images5, and (3) identify all X-ray
ultra-hardband.Thesearethemeanphotonenergiesofde-
sources for exclusion from thestacking backgroundcalcula-
tected sources in the band. Given that bad pixel files for
tions. Source detection was performed using our own pro-
observations in different modes vary considerably, and that
cedure, which is detailed in Nandraet al. 2005. To identify
MERGEALL will only accept one bad pixel file, in order
X-raysourcesforexclusionfromthestackingbackgroundde-
to properly account for their effect we created separate ex- terminationwesetthefalseprobabilitythresholdto10−6,a
posure maps for each of the three groups of observations
levelforwhich1spurioussourceisexpectedperbandinthe
described above. These maps were then re-gridded to pro-
duce effective exposure maps for the full 2Ms. A plot of
∼
solid angle over the BBG survey area as a function of soft 2 Theassumptionthatallcountswillbecontainedwithin30arc-
band effective exposureis shown in Figure 1. secondsmaynotalwaysbestrictlyvalid,particularlyatlargeoff
When performing photometry with Chandra images axis angles (OAAs) and at high energies. However at the mod-
the energy-dependent and position-dependent point-spread erate OAAs considered in this work ( 6 arcminutes) any error
≃
function (PSF) must beproperly accounted for. Dueto the introducedislikelytobeveryminor.
different aim points of the 20 observations in the HDF-N a 3 For more information on the ACIS QE degradation see
givenskypositionwillhaveadifferentPSFineachobserva- http://cxc.harvard.edu/cal/Acis/Cal prods/qeDeg/.
4 To take into account the time dependence of the degradation
tionandthefinalPSFshouldthereforerepresenttheaverage
weappliedtheACISABScorrectiontotheresponsefilesforeach
value.TocalculatethePSFforagivenskypositionweused
individual observation then combined to obtain the exposure-
weightedaveragecorrection.
5 Inordertoincreasesignal-to-noiseratioanyoffsetbetweenco-
1 See http://cxc.harvard.edu/cal/ASPECT/align evt/ for de- ordinatereferenceframesshouldbecorrectedforbeforeperform-
tails. ingstackinganalysis.
4 E. S. Laird et al.
tance less than that of the 95 per cent EEF radius in that
region.Inagivenstackingsamplenormallyzerooroneand
nevermorethan twosources wererejected. Fortheremain-
ing galaxies in the samples we extract counts from the op-
tical positions, summing both the detected counts and the
numberof pixels used for theextraction.
Thebackgroundcountscontributingtothetotalsource
signal are estimated in two ways. First we randomly shuf-
fled the optical galaxy positions by 5–10 arcseconds and
extracted counts from an X-ray source-masked image. Sec-
ondly we used random positions from anywhere over the
10.′3 10.′3 region of interest and again extracted counts
×
from the masked image. In both cases, this was repeated
1000 times for each galaxy – sufficient to get an accurate
Figure 2. Thesignal-to-noiseratiovs.extractionradiusinarc- estimate of the background counts and the dispersion. The
seconds. The signal-to-noise ratio is an inverse measure of the countsandpixelsweresummedthenscaledtothesamearea
fractional error on the flux and is given by (S/√S+B), where as the source extraction. Both background procedures pro-
S and B are the net source counts and background counts re- ducedverysimilarresults–inthispaperwequoteallresults
spectively. The two different background methods, shuffled and for the shuffled positions, which should better account for
random positions, give very similar results. The results using a
any local variations in the background level.
fixedfractionoftheEEFareplottedattheaverageextractionra-
Thestrengthandaccuracyoftheobservedsignalissen-
diusused.Theverticaldashedlinesdenoteourchosenextraction
sitive to the size of extraction radius used. A highly signif-
radiusof1.25arcseconds.
icant soft band detection of a smaller sample of BBGs in
thisfieldhasalreadybeenachievedbyN02:withthedeeper
datawethereforeseek anextractionradiustominimizethe
10.′3 10.′3 surveyarea. To findonly X-raycounterpartsto
× flux errors rather than maximise the significance of the de-
theBBGsthethresholdcanbesetmuchlowerwithoutintro-
tection. We adopted an empirical approach to determining
ducinglikely spuriousX-raysources as thearea in question
theoptimalextractionradius–wetested10fixedextraction
in considerably reduced. We used a threshold of 2 10−4
× radiibetween0.75 and3.0arcseconds,aswellasthe70,80,
which, given the number of BBGs and the area over which
85, 90 and 95 per cent EEF radii (Figure 2), selecting the
the BBG and Chandra catalogues are cross-correlated, re-
optionyieldingthemaximumsignal-to-noiseratio.Wechose
sults in a probability of a false X-ray detection of a BBG a fixed extraction radius of 1.25 arcseconds6, which gave a
of less than 0.15 per cent per detection band. In each case
14.9σ detection with a signal-to-noise ratio of 11.6 (we de-
sources were detected in the full, soft, hard and ultra-hard finedetectionsignificanceasS/√Bandsignal-to-noiseratio
bandsand a band merged catalogue produced.
as [S/√S+B], where S and Bare thenet source and back-
ThelowersignificancebandmergedChandra catalogue
ground counts respectively). The appropriate PSF correc-
was matched totheBBG catalogue to search for anyX-ray
tionsfora1.25arcsecondsextractionradiuswerecalculated
counterparts and to identify any overall shift in the optical
and the flux estimates corrected accordingly. We used the
and X-ray reference frames that should be removed before
full sample of BBGs for the stacking and did not limit the
performing stacking. A small overall offset between thetwo
analysis tosources at small off axis angles.
catalogues was found and the positions of the BBG can-
didates were shifted by -0.47 arcseconds in RA and -0.88
arcseconds in Dec to agree with the reference frame of the
Chandra observation. 3 RESULTS
After correcting for this overall offset we re-matched
3.1 Directly detected sources
thebandmergedChandra cataloguetotheBBGcatalogue,
using a radius of 1.5 arcseconds, to search for X-ray coun- Of the 255 BBGs in our sample 37 are directly detected
terparts. in X-ray. Some of these sources show clear evidence of har-
bouring an AGN (i.e. broad optical emission lines, X-ray
luminosity greater than 1043 erg s−1) while for others the
2.5 Stacking procedure X-ray emission could be due to an AGN, star formation,
or a combination of both. In this paper we are concerned
TodeterminethemeanX-raypropertiesoftheBalmerbreak
with the X-ray properties of galaxies where the emission is
galaxiesthataretooweaktobedirectlydetectedweemploy
likelytobedominatedbynormalstarformingprocessesand
a stacking technique similar to those described in N02 and
we consequently seek to omit from the analysis all galaxies
Brandt et al. (2001).
where AGN emission may dominate. We therefore, conser-
We first identify all the candidates that have an X-ray
vatively,excludeall 37ofthedirectlydetectedgalaxies and
counterpart and exclude them from the stacking. Secondly
weidentifyanycandidateswithanunassociated,nearbyX-
ray source – these objects will be affected by the low-level 6 Wenote that N02contains anerrorregardingthe optimal ex-
PSFwingsoftheChandra PSFandwillleadtoerroneously traction radius used. The paper states that the optimal radius
high source counts. We reject from the stacking all candi- foundwas2.5arcseconds–infact2.5arcsecondswastheoptimal
dates which are separated from an X-ray source by a dis- extractiondiameter.
X-ray properties of star forming galaxies at z ∼1 5
3.3 X-ray and UV correlations for stacked BBGs
The soft band signal for the undetected BBGs is strong
enough to split the galaxy sample into subsets according
to broad band properties and stack the X-ray emission
for several bins, allowing the correlation of X-ray emis-
sion with observed-frame optical magnitudes and colours
to be examined. In order to achieve reasonable signal-to-
noise ratios, for each broad band property we split the
sample into 6 bins based on Un,G,R,I magnitude and
(Un G),(G R),(R I) colour. The range in redshift
− − −
ofthegalaxiesinoursample,∆z 0.6,meansthatagalax-
≃
ies broad-band colour is not just a function of its intrinsic
propertiesbutisalsoafunctionofredshift.Inordertoinves-
tigate any correlations between UV colour and X-ray emis-
Figure3.CountdistributionfortheundetectedBBGsinthesoft sionwithoutacontaminatingredshifteffectwealsocalculate
band.Theverticallinedenotes themeanbackground countsper andbinaccordingtotherelative(Un G),(G R),(R I)
countextractioncell,derivedviatheshufflebackgroundmethod. − − −
colour excess for thegalaxies. For each colour (Un G etc)
−
thecolour excess for each galaxy is calculated from thedif-
ference between the galaxy colour and the median colour
deal only with the undetected BBGs for the remainder of of all galaxies with a redshift within z = z 0.05 of the
±
this paper. At z 1 this equates to removing all galaxies galaxy redshift. These values were binned and the mean
with L2−10 keV &∼1041 erg s−1. Full analysis of the X-ray, colour excess for each bin was shifted so that the lowest
opticalandUVpropertiesofthedetectedgalaxiesincluding valuebin, which presumably contains thebluest and there-
starburstandAGNseparationandidentificationwillbead- foreleastattenuatedgalaxies, wasequaltozero.Wedenote
dressedinanupcomingpaper(Lairdetal.,inpreparation). this shifted colour excess by E(1800˚A–2400˚A), E(2400˚A–
3400˚A)andE(3400˚A–4050˚A) forthe(Un G),(G R)and
(R I) bands respectively, with the wav−elengths−denoting
−
the approximate rest-frame wavelength of the emission in
3.2 Stacking results for undetected sources each filter. The stacking procedureused for the magnitude,
colour and colour excess bins was identical to that carried
Stacking the emission from the undetected BBGs produces
out for thefull samples of undetectedsources.
significantdetectionsinboththesoftandhardbands(Table
TheresultsofthesubsetstackingareshowninFigures4
1).Inthesoftband,stackingthecountsfrom216undetected
(binningaccordingtomagnitude)and5(binningoncolour).
BBGs7 produces a total of 290.8 net counts and a detec-
Foreachquantityweshowtheexactmean0.5–2keVfluxper
tion significance of 14.9. An average of 1.4 net counts per
galaxy in a given bin, with errors, regardless of whether or
galaxy are detected, with 1.76 mean background counts in
not the detection is considered significant. Low significance
each galaxy extractioncell. Thedistributionoftotal counts
detections are easily identifiable by their correspondingly
(S+B) in an extraction cell (Figure 3) covers a wide range,
large error bars. The minimum flux for a 3σ detection is
with an almost continuous distribution of counts from the
peak at 2–3 counts per cell up to 12 counts per cell and f0.5−2keV ≃3×10−18 ergs−1 cm−2(logf0.5−2keV ≃−17.5).
∼ We plot U , G, R and I magnitude versus X-ray flux
the realm of direct detections. Figure 3 demonstrates the n
(Figure 4). Correlations between magnitude and soft X-ray
stackingsignalisnotdominatedbyafewbrightsourcesbut
flux,inthesensethatthebrightestUVsourcesarealso the
instead well representstheentire sample. The detectedsoft
brightestX-raysources,areseen foreach filter.Spearman’s
band signal for the BBGs corresponds to a mean flux per
galaxyof5.51 0.49 10−18 ergs−1 cm−2 (0.5–2 keV)and rankandPearson’s linearcorrelation testsshow eachof the
ameanlumino±sityof×L2−10keV =2.97 0.26 1040 ergs−1. relations to be highly significant (e.g. Pearson’s linear cor-
± × relation coefficients were -0.93, -0.93, -0.95, -0.99, all with
The mean flux per undetected BBG found here is in agree-
ment with that found with the 1 Ms Chandra data for the significance >99 per cent, for Un, G, R and I respectively,
Table2).Rest-frame1–4keVX-rayemissionandrest-frame
smaller sample of BBGs by N02.
UV emission are therefore related even before the effects of
Alowsignificance3.2σ signal isdetectedfrom stacking
dust attenuation are applied. The results of the stacking
the hard band emission from 216 undetected BBGs. There
in the six U bins, which with a rest-frame wavelength of
are87.4totalnethardbandcountsdetected,corresponding n
1800 ˚A at z 1 corresponds to the wavelength commonly
to an average of 0.40 net counts per galaxy. The expected
∼ ∼
used for UV measures of SFR, are shown in Table 1. Only
backgroundlevelis3.4meancountsperextractioncell.This
weak but significant signal results in a mean 2–10 keV flux the weakest bin, 25.25 < Un 6 27.03, does not have a sig-
perBBGof 8.87 2.91 10−18 ergs−1 cm−2.Thespectral nificant (>3σ) X-ray detection.
± ×
constraints implied by this detection are discussed in 4.1 TherelationsbetweenX-rayfluxandthecoloursofthe
§
BBGsislessclear.ForU G,ameasureofthespectralslope
n
−
in the far UV, the X-ray signal of the undetected sources
is approximately flat for varying colour (Figure 5). With
7 Two undetected BBGs are rejected from the stacking as de- respect to G R colour again the X-ray signal is approxi-
−
scribedin 2.6 matelyflat,withperhapssomeevidenceforaslightincrease
§
6 E. S. Laird et al.
Table1.StackingresultsofundetectedBBGsforentiresampleandsub-samplesbasedonrest-frame1800˚Aemission.Col.(1):Galaxy
sample. Col.(2): Observed-frame energy band. Col.(3): Number of galaxies included in stacking sample, taking into account rejected
galaxies as described in 2.5. Col.(4): Mean R magnitude. Col.(5): Mean redshift. Col.(6): Signal-to-noise ratio [S/√S+B], where S
§
and B are the net source and background counts respectively. Col.(7): X-ray flux per galaxy; 0.5–2 keV for soft band, 2–10 keV for
hardband. Col.(8):X-rayluminositypergalaxyinthe2–10 keVband, derivedfromsoftband fluxassumingΓ=2.0andGalactic NH.
10–50 keV luminosity is given for the hard band, also assuming Γ=2.0. Col.(9): SFR from 2–10 keV luminosity (Ranallietal. 2003).
Errorsarestatisticalonly.Col.(10): RatioofX-rayderivedSFRtoUVSFR,uncorrected forattenuation.
Sample Band N hRi hzi S/N (10−18 ergFXcm−2 s−1) (1040LerXgs−1) (MhSFyRr−i1) SFRX/SFRuUnVcor
(1) (2) (3) (4) (5) (6) (7) (8) ⊙(9) (10)
Undetected BBGs Soft 216 23.42 0.95 11.2 5.51 0.49 2.97 0.26 6.0 0.6 3.0 0.4
± ± ± ±
Undetected BBGs Hard 216 23.42 0.95 3.0 8.87 2.91 7.07 2.32 ... ...
± ±
22.57<Un623.72 Soft 34 22.50 0.91 6.7 10.21 1.53 4.81 0.72 9.6 1.4 1.8 0.3
± ± ± ±
23.72<Un624.13 Soft 30 22.96 0.92 5.7 8.81 1.44 3.83 0.68 7.7 1.4 2.6 0.5
± ± ± ±
24.13<Un624.40 Soft 38 23.35 0.97 5.2 6.25 1.21 3.37 0.65 6.7 1.3 2.7 0.5
± ± ± ±
24.40<Un624.71 Soft 40 23.63 0.97 3.1 3.25 1.04 1.75 0.56 3.5 1.1 1.8 0.6
± ± ± ±
24.71<Un625.25 Soft 39 23.83 0.96 3.9 4.02 1.03 2.16 0.56 4.3 1.1 3.2 0.8
± ± ± ±
25.25<Un627.03 Soft 35 24.16 0.97 2.5 2.63 1.07 1.42 0.58 2.8 1.2 4.4 1.8
± ± ± ±
Table 2. Spearman’s rank and linear Pearson’s correlation tests. Col.(1): The two quantities
being tested. Cols. (2) and (3): Spearman rank correlation coefficient and chance probability.
Cols.(4)and(5):LinearPearsoncorrelationcoefficientandchanceprobability.Ineachtestaco-
efficientof-1,0and+1indicatesanegativecorrelation,nocorrelationandapositivecorrelation,
respectively.
Testpair Spearmanrank LinearPearson
Coefficient Prob. Coefficient Prob.
(1) (2) (3) (4) (5)
logfx, Un -0.943 <0.005 -0.927 0.008
logfx, G -0.886 0.019 -0.934 0.006
logfx, R -0.943 <0.005 -0.953 <0.005
logfx, I -1.000 <0.005 -0.988 <0.005
logfx, (Un G) -0.257 0.623 0.143 0.787
logfx, (G −R) 0.428 0.397 0.468 0.349
logfx, (R−I) 0.114 0.623 0.114 0.829
−
logLx, logLUV 0.943 <0.005 0.915 0.010
log(SFRx/SFRUV), E(1800˚A–2400˚A) 0.486 0.329 0.855 0.030
log(SFRx/SFRUV), E(2400˚A–3400˚A) 0.771 0.072 0.892 0.017
log(SFRx/SFRUV), E(3400˚A–4050˚A) 0.714 0.111 0.840 0.036
inX-rayfluxwithincreasingredness.TheresultsforR I, dominated by a few AGN just below the detection thresh-
−
ameasure of thespectral slope neartheBalmer break,also old, as is demonstrated by the counts-in-cell distribution
shownoevidenceofarelationwithX-rayemission.Binning shown in Figure 3. The presence of highly obscured AGN
oncolour excess,which removesanyredshift dependencein can also be ruled out from the observed soft spectrum of
thecolours,producedverysimilar resultstothoseshownin the stacking signal. With significant stacked detections in
Figure 5: UV colour excess therefore also showed no clear boththesoft and hardbandswecalculated themean hard-
correlation withX-rayflux.Spearman’srankandPearson’s ness ratio (HR=(H S)/(H+S), where H and S are the
−
linear correlation tests confirm the absence of a significant net hard and soft band counts) of thesample. Wefindthat
correlation between colour and X-rayflux (Table 2). HR= 0.54 0.11,whichisconsistentwithaΓ 1.8 0.3un-
− ± ∼ ±
absorbedpower-lawspectrumatz 1.Whilelimitedbylarge
∼
errors this result is consistent with the observed spectra of
local starburst and star forming galaxies (e.g. Ptak et al.
4 DISCUSSION
1999) and is inconsistent with the hard spectrum expected
4.1 Stacking signal and SFR estimate if there was a significant population of obscured AGN
within the sample (e.g. the spectrum of the X-ray back-
The X-ray emission from the undetected BBGs included in
ground). This supports the assumption that the emission
the stacking analysis should not be dominated by AGN as
from the BBGs is a result of star formation. In saying this
we excluded from the analysis all galaxies that were di-
rectly detected, equating to a L2−10 keV & 1041 erg s−1 wAeGNsho(uLlLdApGoNin)t, wouitththluamtitnhoesitpyrelseesnsctehaonf loorwcolummpianroasbitlye
cut-off at z 1, therefore removing all moderate and lu-
∼ to the host galaxies themselves, cannot be ruled out. In-
minous AGN. The mean stacking signal is certainly not
X-ray properties of star forming galaxies at z ∼1 7
Figure 4. X-ray stacking results for undetected BBGs split into subsets according to broad band magnitudes. Correlations of soft
X-rayflux with (a) Un magnitude, (b) G magnitude, (c) R magnitude, and (d) I magnitude are shown. At z 1Un,G,R and I filters
correspond to rest-frame 1800 ˚A, 2400 ˚A, 3400 ˚A and 4000 ˚A emission. The x-axis errors are standard∼deviation of values in a
∼ ∼ ∼ ∼
givenbin,andthey-axiserrorbarsarethePoissonerrorsfromstacked softbandcounts.
Figure 5. X-ray stacking results for undetected BBGs split into subsets according to broad band colours. Correlations of soft X-ray
flux with (a) Un G colour, (b) G R colour and (c) R I colour are shown. Un G is a measure of the far-UV slope, G R the
near-UVslopean−dR I thespectr−alslopeneartheBalme−rbreak.Thex-axisandy−-axiserrorbarsareasforFigure4. −
−
deedgiventheobservedincidenceoflowlevelnuclearactiv- therspatially orspectrally,anX-ray–SFRrelation with the
ity in the local Universe (Ho, Filippenko, & Sargent 1997) form of the Ranalli et al. (2003) relation relating the total
it is almost certain that there are more as yet undiscov- 2–10 keV emission to SFR should be used. Only in local
ered AGN within the BBG sample. However the contribu- galaxies, where it is possible to resolve the HMXB popula-
tion of such LLAGN to the mean BBG X-ray luminosity is tion dominating the2–10 keVflux,can theSFRbederived
expected to be small: in a sample of nearby LLAGNs less directly from the total luminosity emitted by HMXBs (e.g.
than one third have X-ray emission that is dominated by Grimm et al. 2003; Persic et al. 2004).
a compact nuclear source and the mean 2–10 keV luminos- An additional possible source of contamination is that
ity of the LLAGNs is less than 1 per cent of that of the of emission from LMXBs which reflects the total stellar
BBGs (Hoet al.2001).Forthegalaxy sampleusedherewe mass of a system as opposed to the instantaneous SFR.
thereforeassumethattheobservedemissionisdominatedby Forgalaxies with lowSFRemission from LMXBscan dom-
star formation processes. For such a sample of galaxies the inated the flux and the X-rays will cease to measure SFR
X-ray emission should be proportional to the SFR allowing (Grimm et al. 2003). While we do not attempt to model
the X-ray luminosity (normally 2–10 keV) to be used as a and separate out the contribution of LMXBs to our mea-
SFRmeasure(N02;Ranalli et al.2003;Grimm et al.2003). sured fluxes we can place approximate upper limits on the
For situations such as the one here where only the total 2– expectedlevelofemissionfromLMXBsinoursample.Stud-
10keVluminosityofagalaxyorsampleofgalaxiesisknown, ies of theLMXB population in theMilky Way give an esti-
andthedatarenderit impossible toresolvetheHMXBsei- mateoftheluminosityfrom LMXBspersolar massofstars
8 E. S. Laird et al.
(Grimm, Gilfanov, & Sunyaev2002). Using this result, and
assuming that the mean stellar mass of the BBGs is no
greater than 1011 M , gives an estimate of the maximum
totalluminosity from⊙LMXBsperBBGof5 1039 erg s−1.
×
Under these conditions the contribution of LMXBs to the
2–10keVluminositywouldthenbe 15percentforthefull
∼
sample,risingto 30percentforthelowestsignificantlyde-
∼
tectedbins.WedonotthereforeexpectLMXBstodominate
theemissionforthissampleofstarforminggalaxies.Spitzer
MIPSdata,aswellasgroundbasedKbandimaging,canbe
usedtoprobethestellarmassesoftheBBGsandwill allow
more stringent estimates of the LMXB contribution to be
made. This will be addressed in futurework.
Given the expected low contribution from AGN and
LMXBs toour detected signal we are confidentthat theX-
Figure6.2–10keVX-rayluminosityvs.1800˚AUVluminosity,
rayemissionfromoursampleofundetectedBBGsisindeed
forbinningonUn magnitude.Solidlineshowsbest-fittingpower
tracing star formation and we proceed according tothisas-
law relation. Dotted line shows best-fitting power law relation
sumption.TheobservedX-rayfluxesandluminositieswhich with1-to-1mappingbetween X-rayandUVluminosity.
correspondtorest-frame1–4keVemissionthereforeofferus
ameasureofSFRatz 1thatisrelativelyunaffectedbyat-
tenuation by dust, for∼column densities of N <1022 cm−2 magnitudes. These correlations, whereby the brightest UV
H
which corresponds to A 50. sources are also the brightest X-ray sources, hold over the
V
Usingtheempiricalre≃lationbetweenthetotal2–10keV rest-frame wavelength range 1800–4000 ˚A. Good, approx-
luminosity and SFR found by Ranalli et al. (2003), we find imately linear, correlations between X-ray luminosity and
that the mean SFR per BBG is 6.0 0.6 M yr−1 (Ta- radio and far infrared (FIR) SFR indicators have been ob-
ble 1), consistent with that found b±y N02 f⊙or a smaller served before for star forming and starburst galaxies, indi-
sample of BBGs using the 1 Ms data. Comparison UV lu- cating that the three wavelength regimes are all following
minosities and estimates of the SFR are calculated from theSFR(Ranalli et al. 2003;David et al.1992).Thecorre-
the flux as measured in the U band, which at z 1 cor- lationseenherebetweenX-rayfluxand1800˚Aemissionalso
n
∼
responds to rest frame 1800 ˚A, and the Kennicutt (1998) impliesthatbothregimesaretracingtheSFR.Asdiscussed
∼
UV luminosity–SFR relation. These luminosities and SFRs above, because rest-frame hard X-rays are relatively unaf-
have not been corrected for dust attenuation and should fected by extinction they should reflect the intrinsic SFR
therefore underestimate the SFR compared to X-ray esti- without the need for extinction corrections. On the face of
mates, which better measure the bolometric value. We find ittheobservedcorrelationwithX-rayisthereforereassuring
SFR /SFRuncor 3 for the full stacking sample, an esti- intermsofusing1800˚AtomeasureSFRathighredshift:in
X UV ∼
mate of the attenuation correction factor for the UV. This general,forUVselectedgalaxiesthesourceswiththeappar-
attenuation factor is slightly smaller than that found by ent highest SFR also appear to be thebrightest in the UV.
Reddy& Steidel (2004) for UV selected galaxies at z 2 HowevergiventhatUVemissionisverysensitivetodustat-
∼
(SFR /SFRuncor 4.5)andbyNandra et al.(2002)forz 3 tenuationthecorrelation isalittlesurprising.Followingthe
LBGsX(SFRU/VSFR∼uncor 5). Thisdifference islikely due∼to generallyacceptedpictureofhigherSFRgalaxiesalsobeing
X UV ∼
the mean SFRs of these samples being significantly larger dustier (e.g. Adelberger & Steidel 2000) one might expect
than that considered here: the attenuation factor is similar the galaxies with large X-ray flux (representing high SFR)
to the factor of 2.3 found for the SFR<20 M yr−1 sub- tohavesuppressedUVflux.InFigure6weshowtherelation
sampleofReddy& Steidel(2004),whichhasa⊙moresimilar betweenX-rayand1800˚AUVluminosityfortheundetected
range in SFRs. galaxies. Spearman’s rank and Pearson’s linear correlation
Itisimportanttoemphasizethatthepropertiesderived tests show a highly significant correlation between log LX
for our sample do not necessarily represent the BBGs as a and log LUV (Table 2). The best-fittingpower-law is
whole.ByexcludingfromouranalysisallBBGswithsignif-
log(L )=(0.62 0.16)log(L )+23.01 4.42 (1)
icant direct X-ray detections we will undoubtedly have re- 2−10 keV ± UV ±
movedfrom thesamplethegalaxies withthehighest SFRs, whichhasanacceptablechi-squared.Thebest-fittoanexact
in addition to the AGN dominated sources. With a cut-off linear slope
luminosity of 1041 erg s−1 this is equivalent to removing
all galaxies wi∼th SFR & 20 M yr−1. The mean SFR, and log(L2−10 keV)=log(LUV)+12.26 (2)
possibly also UVattenuation f⊙actor, of theBBGs as aclass isalsoacceptable,atthe90percentlevel.Withnoevidence
is therefore likely to be higher than the values derived here forincreasingUVattenuationwithincreasingSFR,asmea-
(see 4.2 for further discussion on observed correlations be- sured bytheX-ray,theobserved correlation between X-ray
§
tween SFR,attenuation and UV colour). andUViscounterintuitivetogeneralpictureofhigherSFR
galaxiesbeingdustier.Thesameeffectissuggestedwiththe
SFR /SFRuncor ratios in Table 1: for increasing UV flux
4.2 X-ray–UV correlations and implications X UV
(and following on from Figures 4 and 6, increasing X-ray
The results of the subset stacking in 3.3 (Figure 4) clearly flux and SFR) the mean UV attenuation factors actually
§
showthatthereisacorrelation betweenX-rayfluxandUV decrease (though within the errors the results are also con-
X-ray properties of star forming galaxies at z ∼1 9
Figure 7. Correlations between the ratio of X-ray derived SFR to UV derived SFR (and equivalently A1800) and colour excess for
undetectedBBGs.TheUVSFRwascalculatedfromtheUnbandandisnotcorrectedforattenuation.Thesamplewasbinnedaccording
to (a) E(1800˚A–2400˚A), (b) E(2400˚A–3400˚A) and (c) E(3400˚A–4050˚A) colour excess. Dotted lines show best-fitting relations withthe
formlog(SFRX/SFRUV)=a+(b×colour excess).DashedlinesshowtheCalzetti etal.(2000)attenuation law.
sistent with an approximately constant attenuation factor factors of 1.5 6.5. The observed colour–attenuation re-
∼ −
over the small range in SFRs covered here). These results sults,whilecertainlylimitedbythenumberoffittingpoints
suggestthatUVattenuationmaynotnecessarilybeadirect and large errors, show clearly that UV colour is correlated
function of SFR over therange of SFRsconsidered here. withdustattenuationforUVselectedgalaxiesatz 1.This
∼
Thesubsetstackingalsoshowedtheretobenorelation key result allows a single UV colour to be used to estimate
between redness in the UV and X-ray flux (Figure 5) the UVattenuationcorrectionsandattenuationcorrectedSFRs,
immediate interpretation of which is that there is no direct as shown above in Equation 3.
correlation between the colour of the UV selected galaxies
and their SFRs. Local starbursts exhibit a strong correla- The attenuation coefficients, A1800, and corresponding
tion between UV colour and dust absorption, as measured values of log(SFRX/SFRUV) predicted by the reddening
forexamplebytheratiooffar-infraredtoUVflux,implying curve and attenuation law of Calzetti et al. (2000) are also
that the starbursts redden as the dust absorption increases shown in Figure 7. For a given reddening curve, the im-
and allowing UV attenuation to be estimated purely as a plied relationship between attenuation, A1800, and colour
function of UV colour (MHC99). Furthermore the locally excess, E, is nearly independent of the assumed underly-
derived starburst attention laws appear to hold for high-z ing galaxy template. The Calzetti lines in Figure 7 are ap-
UV-selectedgalaxies suchasLBGs(e.g.Seibert et al.2002; propriate for galaxy spectral types Im, Sc, and Sb which
Reddy& Steidel2004).Inordertoinvestigateifsuchacor- span the expected range for BBGs. While the normalisa-
relation betweencolour andattenuation is seenfor thez 1 tion of our data points with respect to absolute values of
∼
UV selected galaxies and if the local relations hold we plot colour excess was approximate, and therefore the relative
the ratio of X-ray to UV derived SFR for subsets binned normalisation between our data and the Calzetti predic-
on colour excess (Figure 7). Since the X-ray SFR estimates tion is also approximate, themain conclusion is clearly vis-
shouldbeunaffectedbyabsorptionthisratioisameasureof ible: the Calzetti law over predicts the attenuation correc-
UV attenuation. In Figure 7 we also show the correspond- tions compared to the observed values. Between E(1800˚A–
ing attenuation coefficient, A , related to the SFR ratio 2400˚A)=0andE(1800˚A–2400˚A)=1approximately7magni-
1800
bySFR /SFR =100.4A1800.Correlations between colour tudes of extinction are predicted at 1800 ˚A, compared to
X UV
excessandattenuationareseen foreachofthethreebands, the observed difference of 2 magnitudes – an over predic-
∼
withlinearcorrelationtests(Table2)revealingthestrongest tion of the range in UV luminosity and UV SFR correction
correlation to be with E(2400˚A–3400˚A). For each, we fit a factors of 100. For a typical galaxy in our sample with
∼
linearfunctionlog(SFRX/SFRUV)=a+(b colourexcess), E(1800˚A–2400˚A)=0.25,theCalzettilawpredictsA1800 =1.7
×
whereaandbareconstants,whichissimilarinformtothat comparedtotheobservedA1800 =1.1–anoverpredictionin
derived by Calzetti et al. (2000). Each fit yields an accept- theUV SFR correction factor of 1.6. For E(1800˚A–2400˚A),
able χ2, with E(2400˚A–3400˚A) producing the best fit. By E(2400˚A–3400˚A)andE(3400˚A–4050˚A)theprobabilitythat
binning the data on UV colour excess and calculating the the data are derived from the Calzetti model is < 0.01 per
corresponding X-rayandUVSFRsfor each bin wefindthe cent, regardless of the exact position of the Calzetti slope
correlation between UVcolour, UVderivedSFR andX-ray relativetothedatapoints.ThefailureoftheCalzettiatten-
derived SFR to be uation law at predicting the attenuation of the galaxies in
oursampleismost likelyaresult oftheinherentdifferences
SFRX=(1.87+−00..4611)×10(0.44+−00..2205)E(2400−3400)SFRUV, (3) inthesampleofgalaxiesusedhereandthosefromwhichthe
attenuationlawswerecalculated.Theattenuationlawswere
where errors are for ∆χ2 = 2.7, or 90 per cent confidence, derivedforlocalstarburstgalaxiesforwhichtheUVcolours
for one interesting parameter. For the SFRs covered in our arepredominantlyaffectedbydustcontentandobscuration.
sample,andoverarangeincolourexcessof∆E 1,wefind BBG selection on theotherhandidentifies Sb-and Sc-type
≃
that theUV attenuation ranges from 0.5 to2.0 magnitudes starforminggalaxiesaswellasstarbursts(Adelberger et al.
at1800˚A,correspondingtorelativelymodestUVcorrection 2004) and therefore includes galaxies with more quiescent
10 E. S. Laird et al.
star formation, and a range of star formation histories and the support of a PPARC Studentship and a University of
stellarpopulationageswhichaffecttheUVcoloursasmuch London Valerie Myerscough Prize.
asthedustcontent.Suchsamples ofgalaxies clearly donot
follow thestarburst attenuation laws.
REFERENCES
Adelberger K. L., Steidel C. C., 2000, ApJ, 544, 2
5 SUMMARY
AdelbergerK.L.,SteidelC.C.,ShapleyA.E.,HuntM.P.,
With the 2 Ms observation of the HDF-N region we have Erb D. K., ReddyN.A., Pettini M., 2004, ApJ, 607, 266
examined the X-ray properties of a large sample of UV- AlexanderD. M. et al. 2003, AJ, 126, 539
selected, star forming galaxies with 0.74 < z < 1.32. In Bell E. F., 2002, ApJ, 577, 150
this paper we dealt exclusively with the216 galaxies in the Brandt W. N., Hornschemeier A. E., Schneider D. P.,
sample that are not directly detected in X-ray and whose Alexander D. M., Bauer F. E., Garmire G. P., Vignali
emissionshouldbepredominatelyduetostarformationpro- C., 2001, ApJL,558, L5
cesses (analysis of the detected BBGs will be presented in Calzetti D., Kinney A. L., Storchi-Bergmann T., 1994,
an upcoming paper, Laird et al., in preparation). Here we ApJ,429, 582
summarise ourmain results: CalzettiD.,ArmusL.,BohlinR.C.,KinneyA.L.,Koorn-
neef J., Storchi-Bergmann T., 2000, ApJ, 533, 682
(i) StackingthesoftbandfluxfromtheundetectedBBGs David L. P., Jones C., Forman W., 1992, ApJ, 388, 82
resulted in a highly significant, 14.9σ detection. The de- Fabbiano G., 1989, ARA&A,27, 87
tected signal corresponds to a mean flux per galaxy of Giavalisco M., 2002, ARA&A,40, 579
5.51 0.49 10−18 erg s−1 cm−2 (0.5–2 keV) and a mean Grimm H.-J., Gilfanov M., Sunyaev R., 2002, A&A, 391,
lumin±osity o×f L2−10keV =2.97 0.26 1040 erg s−1. 923
± ×
(ii) Stackingthehardbandfluxalso resultedin asignifi- GrimmH.-J.,GilfanovM.,SunyaevR.,2003,MNRAS,339,
cantdetectionoflowlevelemission.Wefindamean2-10keV 793
fluxperundetectedBBGof8.87 2.91 10−18 ergs−1cm−2. HoL.C., FilippenkoA.V.,Sargent W.L. W.,1997, ApJ,
± ×
(iii) Thehardnessratioofthesample,HR= 0.54 0.11, 487, 568
− ±
isconsistentwithaΓ 1.8 0.3unabsorbedpower-lawspec- Ho L. C., et al., 2001, ApJ, 549, L5
∼ ±
trum. This is consistent with the integrated spectrum from Hornschemeier A. E., et al., 2001, ApJ, 554, 742
HMXBs and supports the general interpretation that the HughesD. H., et al., 1998, Nat, 394, 241
detectedemissionisduetonormalstarformationprocesses. Kennicutt R.C., 1998, ARA&A,36, 189
(iv) ThemeanX-rayderivedSFRofthesampleofunde- Lehmer B. D., et al., 2005, AJ, 129, 1
tectedBBGsis6.0 0.6M yr−1.TheratioofX-rayderived Lilly S. J., Le Fevre O., Hammer F., Crampton D., 1996,
SFR to UV derive±d SFR,⊙an indication of the average UV ApJ,460, L1
attenuation factor, is 3. MadauP.,FergusonH.C.,DickinsonM.E.,GiavaliscoM.,
(v) Splitting the sample into subsets based on observed- SteidelC. C., FruchterA., 1996, MNRAS,283, 1388
frameopticalmagnitudesandcoloursandstackingtheX-ray Marshall H.L.,TennantA.,GrantC. E.,HitchcockA.P.,
fluxineachbinproducedsignificantdetections,allowingcor- O’Dell S.L., Plucinsky P. P., 2004, SPIE, 5165, 497
relations between X-ray and UV emission to be examined. MeurerG.R.,HeckmanT.M.,CalzettiD.,1999,ApJ,521,
WefindclearcorrelationsbetweenX-rayfluxandrest-frame 64
UV flux (over the range 1800–4000 ˚A) such that galaxies Nandra K., Mushotzky R. F., Arnaud K., Steidel C. C.,
∼
thatarebrighterintheUVarealsobrighterinX-ray.There AdelbergerK.L.,GardnerJ.P.,TeplitzH.I.,Windhorst
is no evidence for a correlation between UV colour or UV R.A., 2002, ApJ, 576, 625
colour excess and X-ray flux. NandraK., et al., 2005, MNRAS,356, 568
(vi) Correlations between colour excess and the ratio of Persic M., Rephaeli Y., Braito V., Cappi M., Della Ceca
X-ray-to-UV SFRs (which measures UV attenuation) are R.,Franceschini A., Gruber D.E., 2004, A&A,419, 849
seen, whereby the ratio of X-ray-to-UV increases with UV Ptak A., Serlemitsos P., Yaqoob T., Mushotzky R., 1999,
redness. Using the data we derive a relation for estimating ApJS,120, 179
UV attenuation corrections from UV colour excess. Ranalli P., Comastri A., Setti G., 2003, A&A,399, 39
(vii) Theobservedrelationbetweencolourexcessandat- ReddyN. A., SteidelC. C., 2004, ApJ, 603, L13
tenuationisfoundtobeinconsistentwiththeCalzetti et al. Seibert M., HeckmanT. M., Meurer G.R., 2002, AJ, 124,
(2000) and MHC99 extinction curve which over predict the 46
rangeinattenuationcorrectionsforthegalaxiesinthissam- StarkA.A.,GammieC.F.,Wilson R.W.,Bally J., Linke
ple bya factor of 100. R.A., Heiles C., Hurwitz M., 1992, ApJS,79, 77
∼
Steidel C. C., Giavalisco M., Pettini M., Dickinson M.,
AdelbergerK. L., 1996, ApJ,462, L17
Steidel C. C., Adelberger K. L., Giavalisco M., Dickinson
ACKNOWLEDGMENTS
M., Pettini M., 1999, ApJ, 519, 1
We thank the referee for helpful comments that improved SteidelC.C.,AdelbergerK.L.,ShapleyA.E.,PettiniM.,
themanuscriptandtheChandra X-rayCenterforassistance Dickinson M., Giavalisco M., 2003, ApJ, 592, 728
with some analysis and software issues. D. Rosa-Gonz´alez SteidelC.C.,ShapleyA.E.,PettiniM.,AdelbergerK.L.,
is thanked for many useful discussions. ESL acknowledges ErbD.K.,ReddyN.A.,HuntM.P.,2004,ApJ,604,534